VNC walking CPG

The most rigorously validated project in this batch, and the rating is capped by the scale's own definition rather than by the work's quality. The model is a firing-rate ODE over the VNC connectome - dR/dt = (half-tanh(a/fr_cap * (input - threshold)) * fr_cap - R) / tau, per-neuron tau, threshold, gain and firing-rate cap drawn from a stated distribution, integrated with an adaptive-step Dopri5 solver - so neuron state changes under a documented model: L1. Not L2, and this is the decisive point: no synapse is ever updated. The connectivity matrix is scaled once at setup by fixed excitatory/inhibitory multipliers, optionally perturbed with static noise or pruned for a robustness sweep, and then held constant for the whole integration; a grep for plasticity, STDP, dopamine, learning rate or eligibility across every .py and .yaml in the repo returns nothing. The preprint's own abstract states that the simulation predicted a DNb08 rhythmic-walking pathway that was then confirmed experimentally. If the site wants to recognise validated non-plastic circuit models, this project is the case that forces the question. It carries the validation mark, and it is the project that forced the scale to change: the authors activated DNg100 and DNb08 optogenetically in real tethered flies, tracked leg kinematics with Anipose, and scored those recordings with the same rhythmicity metric they apply to simulated neurons — with the simulation predicting the result before the experiment confirmed it. A cumulative scale filed that at L1 next to a sandbox.